2023-04-26 16:52:09.277957: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/repli

这个错误是由于在 `MultiHeadAttention` 层中，`axis` 参数设置不正确导致的。默认情况下，`axis=-1` 表示对最后一个轴进行 softmax 操作，但是在你的代码中，输入数据的最后一个轴是时间步，而不是特征维度。 因此，你需要将 `axis` 参数设置为正确的值，即 `axis=2`，表示对特征维度进行 softmax 操作。修改代码如下： ```python from keras.layers import Input, Dense, CuDNNGRU, Dropout, ActivityRegularization from keras.layers import Concatenate, Flatten, Reshape, Lambda from keras.layers import LayerNormalization, MultiHeadAttention def LSTNet(trainX1, trainX2, trainY, config): input1 = Input(shape=(trainX1.shape[1], trainX1.shape[2])) input2 = Input(shape=(trainX2.shape[1], trainX2.shape[2])) # CNN layer for temporal feature extraction conv1 = Conv1D(filters=config['conv_filters'], kernel_size=config['conv_kernel'], activation='relu')(input1) dropout1 = Dropout(config['dropout'])(conv1) # GRU layer for capturing temporal dependencies gru1 = CuDNNGRU(units=config['gru_units'], return_sequences=True)(dropout1) dropout2 = Dropout(config['dropout'])(gru1) # Attention layer for capturing temporal dependencies across multiple time series attention1_output = MultiHeadAttention(num_heads=4, key_dim=16, attention_axes=(1, 2))(dropout2, dropout2) attention1_output = LayerNormalization()(attention1_output) dropout3 = Dropout(config['dropout'])(attention1_output) # Concatenate with external features concat1 = Concatenate(axis=2)([dropout3, input2]) # Fully connected layer for prediction flatten1 = Flatten()(concat1) dense1 = Dense(units=config['dense_units'], activation='relu')(flatten1) dropout4 = Dropout(config['dropout'])(dense1) output = Dense(units=1, activation='sigmoid', activity_regularizer=ActivityRegularization(l1=config['l1'], l2=config['l